Cleaning decision model of MBR membrane based on Bandelet neural network optimized by improved Bat algorithm

Abstract

The membrane fouling is an important factor of restricting wide application of MBR (Membrane Bio-Reactor), which causes the fall of membrane flux and reduces the membrane cleaning period. So the Bandelet neural network is proposed through combining Bandelet transform and neural network, which predicts membrane flux and its recovery rate for making proper membrane cleaning decision. Firstly, the main affecting factors of membrane fouling are discussed. Secondly, the architecture of Bandelet neural network is designed with Bandelet function and its scale function as activation functions of hidden and output layers respectively. Thirdly, the improved Bat algorithm is established, which is applied to improve the optimization effect of parameters of Bandelet neural network. Finally, the simulation analysis is carried out, the improved bat algorithm has higher performance than the traditional bat algorithm through analyzing the single objective optimization problem from 2018 CEC competition, the optimal number of nodes in hidden layer is confirmed based on comparison analysis and statistical tests. The proposed BNN-IBA has obvious superiority in prediction accuracy and speed according to prediction simulation results of membrane fouling of MBR, which has better prediction results than other state-of-art prediction models optimized by the novel optimal algorithms. In addition, the proper membrane cleaning period and method are confirmed according to the prediction results of membrane flux and its recovery rate.